DEGnext

DEGnext predicts significant upregulating (UR) and downregulating (DR) genes from RNA-seq cancer datasets using a Convolutional Neural Network (CNN) with transfer learning to improve differential expression analysis for small sample sizes.

Key Features:

• Deep Learning Integration: Uses a Convolutional Neural Network (CNN) to capture complex patterns in high-dimensional gene expression data.
• Transfer Learning Application: Applies transfer learning to transfer trained feature maps to novel cancer datasets and improve prediction on small sample sizes.
• Biologically Validated Data Utilization: Incorporates biologically validated data and logarithmic fold change values to classify differentially expressed genes as UR or DR.
• Competitive Performance: Achieves Receiver Operating Characteristic (ROC) scores of 88–99% and outperforms Decision Trees, K-Nearest Neighbors, Random Forests, Support Vector Machines, and XGBoost in UR/DR classification.
• Robustness Across Datasets: Demonstrates effective classification of unseen datasets through transferred feature maps across cancer types.
• Biological Validation: Validates predicted DEGs against significant Gene Ontology terms and cancer-related pathways.

Scientific Applications:

• Biomarker Discovery in Oncology: Identifies potential biomarkers by classifying differentially expressed genes in cancer RNA-seq datasets.
• Cross-Cancer Gene Expression Analysis: Generalizes learned patterns across datasets to explore gene expression changes associated with various cancers.
• Extension to Other Diseases: Can be applied to other disease datasets for differential expression classification using transferred feature maps.

Methodology:

Implements a Convolutional Neural Network (CNN) with transfer learning using biologically validated data and logarithmic fold change values to classify genes as UR or DR; evaluates performance by Receiver Operating Characteristic (ROC) and compares to Decision Trees, K-Nearest Neighbors, Random Forests, Support Vector Machines, and XGBoost; validates predicted DEGs against Gene Ontology terms and cancer-related pathways.